OPHI RESEARCH IN PROGRESS SERIES 36c

Transcription

1 Oxford Poverty & Human Development Initiative (OPHI) Oxford Department of International Development Queen Elizabeth House (QEH), University of Oxford OPHI RESEARCH IN PROGRESS SERIES 36c Multidimensional poverty measurement for EUSILC countries Sabina Alkire 1, Mauricio Apablaza 2 and Euijin Jung 3 Revised Draft, September 2014 Abstract This paper presents a set of experimental indices of multidimensional poverty, using crosssectional EU SILC data. The indices use the Alkire Foster (AF) methodology a widelyused flexible methodology which can accommodate different indicators, weights and cutoffs. In constructing three sets of illustrative indices we review the joint distribution within and among potential indicators of multidimensional poverty. We draw on existing EU2020 indicators, as well as on indicators of health, education and the living environment. The time series data enable an analysis of multidimensional poverty dynamics, including analyses of changes in overall poverty and in indicators. The paper also decomposes poverty results by gender finding women to be poorer across countries and time and by age categories. Keywords: Multidimensional poverty, Counting measures, Material deprivation, Multidimensional poverty dynamics JEL classification: I3, I32, D63 1 University of Oxford, United Kingdom. 2 Facultad de Gobierno, Universidad del Desarrollo (Chile) and the University of Oxford (UK). 3 Korea Research Institute for Vocational Education & Training (KRIVET). OPHI gratefully acknowledges support from the UK Economic and Social Research Council (ESRC)/(DFID) Joint Scheme, Robertson Foundation, Praus, UNICEF N Djamena Chad Country Office, German Federal Ministry for Economic Cooperation and Development (BMZ), GeorgAugustUniversität Göttingen, International Food Policy Research Institute (IFPRI), John Fell Oxford University Press (OUP) Research Fund, United Nations Development Programme (UNDP) Human Development Report Office, national UNDP and UNICEF offices, and private benefactors. International Development Research Council (IDRC) of Canada, Canadian International Development Agency (CIDA), UK Department of International Development (DFID), and AusAID are also recognised for their past support.

2 Acknowledgements: We are grateful to Tony Atkinson, Anne Catherine Guio, James Foster, Bertrand Maitre, Eric Marlier, Brian Nolan, and Nicholas Ruiz for insightful conversations and/or technical support, to Saite Lu and Garima Sahai for research assistance. All errors remain our own. This work has been supported by the second Network for the analysis of EUSILC (NetSILC2), funded by Eurostat. The European Commission bears no responsibility for the analyses and conclusions, which are solely those of the authors. Corresponding author: Citation: Alkire, S., Apablaza, M., and Jung, E. (2014). Multidimensional Poverty Measurement for EU SILC Countries. OPHI Research in Progress 36c, Oxford University. This paper is part of the Oxford Poverty and Human Development Initiative s Research in Progress (RP) series. These are preliminary documents posted online to stimulate discussion and critical comment. The series number and letter identify each version (i.e. paper RP1a after revision will be posted as RP1b) for citation. For more information, see

3 Introduction Methodologies of multidimensional poverty measurement that draw on the counting approach have been used in policy applications since the 1970s (Townsend 1979; see also Atkinson 2003, Nolan and Whelan 2011, and Alkire et al Ch 4 for reviews), and are gaining greater momentum (Erikson 1993, Callan et al. 1999, Atkinson 2003, Chakravarty and D Ambrosio 2006, Whelan et al. 2014). To date many studies have focused on understanding the structure among deprivations, and on identifying the normative, policy, and statistical tools that can best justify the collection of data on distinct indicators (Atkinson et al. 2002; Atkinson et al. 2005, Atkinson and Marlier 2010b and the references therein). Others have used statistical methods to justify why indicators might be aggregated into a composite indicate covering one relevant dimension such as material deprivation (Guio et al. 2012; OECD 2008). Drawing upon such studies, this paper presents a set of experimental indices of multidimensional poverty which use an adjusted headcount ratio M 0 that builds on a countingbased dualcutoff methodology (Alkire and Foster 2011a, 2011b). We show how these measures can be used to provide diverse and specific descriptive analyses, hence why they may complement existing measurement approaches. The methodology is flexible in that different indicators, cutoffs and weights can be used, including cardinal, ratioscale, binary, ordinal and categorical variables. Unlike the headcount ratio which has been traditionally used with countingbased measures in both Europe and Latin America, the AF family of measures incorporate the joint distribution of deprivation and include a new feature of intensity which shows the percentage of dimensions in which the average poor person is deprived. Incorporating intensity into the measure itself enables the multidimensional poverty measure to be broken down by indicator (after identification), to show the levels and composition of deprivations poor people experience. This is not possible with countingbased headcount ratios. Measured poverty also changes if intensity changes. Where data permit, the measure and each of its consistent indicators can be further broken down by subgroups such as gender, age, social groups or regions. The global Multidimensional Poverty Index (MPI) which is released by UNDP s Human Development Reports and covers 108 countries in 2014 is based on this methodology (Alkire and Santos 2010; UNDP 2010), as are official national measures of multidimensional poverty, such as those of Mexico, Colombia, the Philippines, and Bhutan. 1

4 The first application of the AF method in OECD countries, was implemented using the 2009 EU SILC dataset by Whelan Nolan and Maitre in This paper extends Whelan et al. s work by constructing AF poverty measures across time periods , using, necessarily, a more limited set of indicators. In doing so, we demonstrate the analysis of the multidimensional poverty indicator in one period and across time, by headcount, intensity, and indicator. The contribution of this paper is to show the kinds of policy analyses that could be done using this methodology, were a set of dimensions and indicators to be agreed upon by a legitimate process, and were fully consistent and comparable variable definitions to be used. The paper proceeds as follows. Section 1 briefly situates our topic in the literature and Section 2 introduces the AF methodology. Section 3 introduces the data then presents three experimental indices of multidimensional poverty, using crosssectional EUSILC data and the individual as unit of analysis. It first presents the nonresponse and longitudinal availability of information and describes the deprivations in each indicator ( uncensored headcount ratios ) for each country, then explores associations across indicators, and three weighting structures. Section 4 presents the AF results, first showing the poverty cutoff for each decile across time and across measures to illustrate the likely robustness of analyses. Choosing a poverty cutoff for each measure, it then presents the overall results of the three measures across all countries having data in all periods and their component partial indices: the headcount ratio or percentage of the population identified as multidimensionally poor (H), and the intensity, or average percentage of weighted deprivations experienced by poor people (A), and censored headcount ratios. Censored headcount ratios show the percentage of people who are identified as poor and are deprived in each particular indicator. All three measures show a significant reduction in poverty , although patterns vary. Across dimensions, the composition of poverty varies across the three measures according to the weights. Section 4.2 illustrates the level and composition of poverty by countries in 2009, and also compares uncensored and censored headcount ratios on each indicator to see which are most highly associated with poverty. Section 4.3 presents annual time comparisons on aggregate and by country, and assesses the significance of changes over time. As the measure uses individual level data, Section 4.4 decomposes results by gender and age group, finding women to be poorer across time and space. Section 5 concludes. 2

5 1. Motivation Multidimensional approaches to poverty and deprivation have a long and distinguished history in conceptual and philosophical work (Sen 1992). In terms of policy, the late 1960s and early 1970s saw the entrance of policy applications, with the 1968 Swedish Level of Living Study (Johannson 1973, Allardt and Uusitalo 1972); Jacques Delors 1971 Les indicateurs sociaux and P.Ch. Ludz s Materialien zum Bericht zur Lage der Nation (1971), each providing independent impetus in different countries and across Europe for this effort. In more recent literature, significant attention has been paid to the relationship among deprivations, to ways of communicating these, and to methodologies to validate indicators used in composite or multidimensional indices (Atkinson et al. 2002; Alkire et al. 2015, Callan et al 1993; Gordon et al 2003; Layte et al 2001, Nolan and Whelan 1996, 2010, 2011, OECD 2008, Saunders and Adelman 2006; Whelan 2007). Drawing on the 2004 EUSILC data, Guio and Maquet (2006) proposed a multidimensional indicator of Material Deprivation, which reflected deprivations such as poor housing, lack of durable assets, and an inability to afford to meet basic needs. The indicator was designed to be comparable across time and across the EU and most member states, and to provide meaningful trend data showing improvements in material deprivation over time. Whelan (2007) used the Irish component of the 2004 EUSILC dataset to develop an 11item consistent poverty index; and Whelan and Maître (2009) use a range of statistical methods such as correlation and factor analysis; goodness of fit tests like root mean square error of approximation; and reliability tests like Cronbach s Alpha, to identify three dimensions of material deprivation (consumption, household facilities, and neighbourhood environment) and examine their relationship to income poverty. Coromaldi and Zoli (2012) clarify the added value of nonlinear principal component analysis, NLPCA, to these techniques. Guio et al. (2012) provide a systematic exposition of an expanded range of techniques to justify a new severe material deprivation index using the 2009 EUSILC dataset. A set of parallel papers explores similar questions with respect to child poverty (Bradshaw 2009, Notten and Roelen 2010, Gabos et al 2011, Guio et al 2012, and Adamson 2012). Naturally, this deep analysis of the structure of deprivations resulted in a set of empirical and policy studies on the relationship between income and other deprivations (Verbist and Lefebure 2008, Whelan and Maitre 2009, Jana et al 2012) and also gave rise to applied multidimensional measures (Whelan et al 2014). 3

6 The EUSILC dataset has also been used by academic studies to illustrate multidimensional poverty measurement methodologies (Chakravarty and D Ambrosio 2006; Bossert, et al 2013, among others). Brandolini (2007) explored Atkinson s (2003) counting approach using data for France, Germany and Italy and a headcount ratio associated with the minimum proportion of deprivations a person has, and compared the various deprivation measures with income poverty measures. He drew attention to the sensitivity of crossnational comparisons to weights, and also to the deprivation cutoff. This paper adds to this already significant recent literature by illustrating the rich variety of analyses that can be accomplished using one particular methodology, drawing on three experimental measures which differ in indicator weights. 2. AF Methodology This section briefly introduces the class of M α measures developed by Alkire and Foster (AF) that build on the Foster Greer Thorbecke (FGT) index, using the notation found in other works (Alkire and Foster 2011a). The three experimental measures use the M 0 methodology in this class. There are a total of n persons and the wellbeing of each is measured in a total of d dimensions. When referring to a particular person we call them i, and we call a particular dimension, j. The whole dataset is collected in a matrix, where row i tells us the achievement for person i on each of the different dimensions j from 1 to d, and where column j tells us the score on dimension j of each person i from 1 to n. So looking across a row of the matrix gives the full picture for one person, and looking down a column gives the full picture for a given dimension. In weighting dimensions (not people) we use weights w j where these sum to 1. For each column of the matrix, we set a cutoff, z j, for that dimension of deprivation. We then construct a deprivation matrix!! by going down each column, and setting the entry for person i equal to w j if they are below the cutoff. Where they are at the cutoff or above the entry is set to zero. So a person who is not deprived on any dimension has a row of zeroes. For each person we now look at the row and add up the positive entries. This gives us a new column, with entries c i where i goes from 1 to n. This is the sum of the weighted deprivations suffered by person i. We call the column vector of c i entries the count vector, and each entry c i show s a person s weighted deprivation score. 4

7 Next, we identify who is multidimensionally poor. A poverty cutoff k is selected whose value is greater than zero but less than one, and is applied across column vector c. A person is identified as poor if their weighted deprivation score!!!. For example, if a person is deprived in 40% of the dimensions (that is their weighted deprivation score is 40%) and the poverty cutoff is 20%, that person is identified as poor because 40% > 20%. This can be called a dual cutoff identification method, because it uses the deprivation cutoffs!! to determine whether a person is deprived or not in each dimension, and the poverty cutoff k to determine who is to be considered multidimensionally poor. 1 Having identified the poor, construct a censored deprivation matrix!!!, obtained from!! by replacing its!!! row!!! with a vector of zeros whenever!! <!. This censored deprivation matrix contains the weighted deprivations of those persons who have been identified as poor and replaces deprivations of the nonpoor with zeros. The censored deprivation matrix is the basis of the dimensional partial indices. For example, the censored headcount ratios are simply the mean of its columns, divided by the weight of that column. The measure M 0 is the mean of the censored vector of deprivation scores (c i (k)). 2 M 0 can also be expressed as the product of the (multidimensional) headcount ratio (H) and the average deprivation share among the poor (A). H is simply the proportion of people that are poor, or q/n where q is the number of poor people. A is the average share of weighted deprivations poor! people experience! =!!!!! (!)! and reflects the average intensity of multidimensional poverty. M 0 satisfies a number of useful axioms, specifically: replication invariance, symmetry, poverty focus, deprivation focus, weak monotonicity, nontriviality, normalisation, dimensional monotonicity, subgroup decomposability, dimensional breakdown, ordinality and weak rearrangement (Alkire and Foster 2011a, 2013). These axioms are joint restrictions on the methodology that includes both identification and aggregation steps. If data are cardinal, other measures within the M α family can be computed. These measures can reflect the depth and severity of multidimensional poverty, and satisfy 1 This identification strategy can also be represented, following Bourguignon and Chakravarty (2003), by an identification function!: R!!!!! R!! {0,1}, which maps from person i s achievement vector!! R! and cutoff vector z in R!! to an indicator variable in such a way that!!! ;! = 1 if person i is poor and!!! ;! = 0 if person i is not poor. 2 M 0 is the mean of the matrix when the weights sum to d. In this notation, because weights sum to 1, M 0 is the mean of the matrix multiplied by the number of columns or dimensions d. 5

8 other axioms related to monotonicity and transfer. However these are beyond the scope of this paper because most of the EUSILC variables are not cardinally meaningful. For tracking changes across time, different approaches are possible. Naturally the number, level and significance of changes in poverty measures and their associated partial indices can be directly compared, and absolute and relative rates of change can be analysed. Alkire et al Ch 9 provides a systematic presentation of different methodologies for assessing poverty dynamics using repeated crosssection data.3 3. Data and Indicators In 2001, the Laeken European Council endorsed a set of 18 indicators of social inclusion for Europe which were subsequently refined, consolidated and extended, using normative, statistical, and policy reasoning. Atkinson et al. (2005) traces how this process led to the agreement of common social indicators related to deprivation, housing and services, which in turn gave rise to common survey instruments. The European Union Statistics on Income and Living Conditions (EUSILC) was developed precisely to compare deprivation and social exclusion across European countries. Data are available annually for most countries, with the earliest data being available from 2005, and other countries being added gradually. Atkinson and Marlier (2010b) provide an overview of the survey initiation (Chapter 2, Figure 2.1). The datasets provide harmonized information needed to assess being atriskof[income]poverty as well as indicators such as (quasi) joblessness, health, housing and the lived environment. This paper selects an illustrative set of 12 indicators and compares three measures made from these indicators across time and space. It is important to note that the illustrative measures are limited by variable definition (comparable variables must be present across time periods and must be accurate at the unit level rather than only on average) as well as by data availability (missing values in any variable must be low). Particular challenges are evident in the educational data because the years of schooling 3 If the strong assumptions underlying theoretical decompositions required can be justified, Shapley value decompositions (Roche 2013) and other decompositions (Apablaza and Yalonetzky 2011) can be used to explore the percentage of poverty reduction which can be attributed to a reduction in headcount vs. intensity, and by indicator and demographic changes. However the required assumptions for either approach are difficult to justify empirically (Alkire Roche Vaz 2014). 6

9 that correspond to primary education vary across EUSILC countries as may educational quality. Also, data for some indicators including the health variables are subjective or selfreport, and may not accurately proxy the level or trend of objective outcomes. For income poverty and material deprivation our indicators are constructed like the EU2020 multidimensional poverty measure component indicators 4. The lack of detailed information regarding parttime jobs before 2009 renders unnecessary the precise replication of the EU2020 quasijoblessness indicator, but does provide comparability across years for a similar indicator. In our (quasi) joblessness indicator, we constructed a quasijoblessness condition considering all members of relevant households. Households that exclusively contained any one of the following three groups: children 018, students in selfdefined current economic status 1824, or persons aged 60 and above; were considered nondeprived in (quasi) joblessness. Also, because of data limitations we are not able to implement the 2009 severe material deprivation index with improved indicators proposed in (Guio et al. 2012), nor to replicate Whelan et al. (2014) multidimensional poverty measures, because both draw on variables that are present from 2009 but not in previous periods. Where aggregate figures are reported, these include information only from countries with data available across all years. The aggregate figures include (populationweighted) data of Austria, Cyprus, Czech Republic, Denmark, Estonia, France, Germany, Greece, Hungary, Iceland, Italy, Latvia, Lithuania, Luxembourg, Netherlands, Norway, Poland, Portugal, Slovakia, Slovenia, Spain, Sweden and United Kingdom. We excluded from the aggregate results information of Belgium (2012), Bulgaria (2006), Croatia ( ), Ireland (2012), Malta ( ), Romania (2006) and Switzerland ( ). Additionally, due to irregularities in the education variable PE040 (Highest ISCED level attained), Finland (2007) was also excluded. Finland shows that all individuals have primary education across all years except in In 2007, 18% of the populations did not have primary education. For national results, we include all countries 5. 4 To measure severe material deprivation we identify households as deprived if they have 4 or more deprivations in the 9 indicators used in the EU2020 measure (listed in Table 5). 5 We also observed uncommon changes in housing in Hungary (2008) and Bulgaria ( ) and unmet Medical Needs in Portugal (2007) but numbers were contrasted and corroborated with official statistics. For proof see and for unmet Medical Needs see usion/indicators/health_long_term_care 7

10 Regarding the sample weights, a subset of countries (Denmark, Finland, Iceland, Netherlands, Norway Slovenia and Sweden) have a lower sample size for health variables due to their use of registry data. MPIs are calculated only for respondents having all indicators, and aggregated using the specific weight for this subselected population (PB060). To maintain the comparability of the aggregate and pooled results we reweighted the data using the retained sample for each country (dropping missing values as reported in Table 1) to correspond to the number of individuals of each country as in the original dataset. Regarding the analysis over time, point estimates calculations are complemented with the analysis of standard errors and tests of differences of mean, following the structure proposed by Goedemé (2010, 2013). 3.1 Unit of Analysis: Individuals 16 years and above Different units of identification are possible using the EUSILC dataset: individual adults; individual children; adults or children by household, and households. 6 Here we use the individual adults as a unit of identification. The measures that follow combine individual and household level information, and identify each individual aged 16 and above as multidimensionally poor or nonpoor based on his or her own achievements in the health and education indicators for which this information is available. Household level variables are used to identify individuals as deprived or nondeprived in terms of at risk of income poverty, severe material deprivations, (quasi) joblessness, housing, noise, crime and pollution. This way of proceeding is useful because the resulting measures can be disaggregated by gender and age. However normatively iusing the individual adult as a unit of identification overlooks (and does not foster) intrahousehold sharing and caring in the individually measured dimensions. For example having a chronic disability in a household which can effectively care for such a person is very different than having the same health condition and living alone. Nor does it capture the possibility of externalities in the household such as an educated member helping another. Some policy aims support a household focus. It would be possible to use the household as a unit of identification for a poverty measure built with EUSILC data. In this case, a household would be deprived in education, and health indicators depending upon the joint deprivations of all household members (which might include children) for whom data were available. This method which was used for example in the global MPI can reflect 6 For an extensive discussion of the household structure in Europe, see Chapter 4 by Maria Iacovou and Alexandra Skew in Income and living conditions in Europe (Atkinson and Marlier, 2010). 8

11 intrahousehold sharing and child deprivations. In this case, the results still can be aggregated using individual sampling weights such that the unit of analysis (individual) reflects the proportion of people who are poor (unit of analysis is individual) or the proportional of households who are poor (unit of analysis is the household). Here, the household was not used as the unit of identification or analysis in these measures, in part because household structures vary across Europe (Iacovou and Skew 2010). Also, the appropriate cutoff for household level indicators built with individual education and health data would require separate analysis. 7 Finally, in the EUcontext, social rights tend to be individually based. For that reason, in the experimental measures the individual is taken as a unit of identification, with the consequence of not including child poverty. 3.2 Dimensions, Indicators, and Weights The dimensions and indicators of deprivation in this paper draw upon an earlier paper in which we implemented four experimental measures, each having three to six dimensions and a variety of differently defined indicators, using data from (Alkire Apablaza and Jung 2012), as well as an interim draft in which we implemented further experimental measures, including one that omitted chronic health conditions. Based on these results, and the editors and others comments upon them, we revised the dimensions and indicator definitions a third time. The three experimental indices presented in this paper have four, five and six dimensions. They use nested weights in that each dimension is equally weighted, and each indicator within a dimension is likewise equally weighted. Dimensions of health and education and some form of economic welfare are present in most descriptions of multidimensional poverty (Appendix 1, Alkire 2002). Drawing on the arguments provided in Whelan et al (2014) and Guio and Maquet (2006), all measures adds to these a dimension of the living environment, which includes housing and neighbourhood considerations: noise, pollution and safety. In all measures, education, health, and the living environment each enter as separate dimensions. In Measure 1, the EU2020 indicators together form the fourth dimension. In Measure 2, 7 The aggregation of intrahousehold data and the setting of deprivation cutoffs require normative, policy, and empirical exploration to justify. They can be set based on a counting approach across household members or some alternative aggregation. For example, a household can be considered deprived in education, if a) one household member has not attained a certain educational level; b) no one in the household has attained a certain educational level; c) at least onethird of household members have not attained a certain level, or d) if the average achievement level across household members is less than some threshold. Of course, households differ in kind as well as by cultural or geographical group: nuclear or extended families differ from student houses and migrant workers sharing accommodation, and the assumptions of intrahousehold sharing must be considered for distinct household types (Alkire and Santos 2014). 9

12 AROP and quasijoblessness form one dimension and material deprivation is a separate fifth dimension. In Measure 3, each indicator of the EU2020 poverty index becomes its own separate dimension making a total of six dimensions. Effectively, this changes the relative weight of different poverty components in Measure 1, EU2020 indicators contribute 25% of the total weight; in Measure 2, 40% and in Measure 3 it rises to 50%. Taken together a comparison across these measures also illustrates the robustness of the analysis to changes in weights. Terminologically, dimensions are organising concepts which in this case govern the weights attached to indicators. They may also be used to communicate the results in public. Once again, the discussion of the appropriate dimensions to organise the measurement of deprivation has a long history, which can inform present discussions. Because these measures are experimental we do not provide an extensive normative justification of the dimensions drawing on people s own values, the theoretical literature, the policy purpose of the measure, and other considerations. Such an extensive justification is provided in the case of official multidimensional poverty measures. Appendix one provides a set of dimensions and in some cases indicators that have been used in the European context (see also Atkinson et al 2002). The indicators of these measures are data constrained. EUSILC indicators tend to be defined in the space of resources, in the case of AROP, severe material deprivation or housing or common proxies for functionings, such as levels of schooling and employment status. Some draw upon selfassessments for example, evaluations of noise and safety and health which may not reflect the objective risk of violence or noise vibrations in a neighbourhood. If a measure is intended to reflect deprivations in the functionings or capabilities that poor people experience (Sen 1992), then it would be necessary to examine in what way each indicator could be interpreted to proxy functionings and the anticipated accuracy of such proxies for diverse individuals. Rather than doing so, in this case we draw upon the rich existing literature justifying the EUSILC indicators (Atkinson and Marlier 2010). Table 5 describes each component indicator of the experimental measures and its deprivation cutoff. Several notes may be in order. First, other studies have not necessarily included education, perhaps due to country differences in the definition of levels of education. These measures retain education because of its importance, and consider a person to be deprived if they have not completed primary school. But the indicator is not necessarily comparable, because the same levels of education may correspond to differing number of years in different countries. 10

13 In terms of the severe material deprivation indicator, Guio et al. (2009) and subsequent work had used a cutoff of three out of nine indicators to signify material deprivation. In more recent work (Guio et al. 2012, Atkinson et al 2010), both the variables and the cutoffs have been reassessed using the 2009 EUSILC dataset. In this paper we follow Guio et al. s (2012) stricter version of the indicator severe material deprivation with a cutoff to four out of nine using the original Guio et al (2009) indicators. In employment, we had to modify the (quasi) joblessness indicator proposed by EU2020 to apply to the entire population. Although ours follows the same structure8, we identify as nondeprived the people not included in the original indicator: children (<18), students between 18 and 24 and elderly (>60). The AtRiskOf[Income]Poverty indicator follows the EU2020 standards, and considers a person at risk of poverty (AROP) if their household income is less than 60% of the national median equivalised disposable income. 3.3 NonResponse In this analysis, we have adopted a rigorous approach to the issue of missing values. At the country level we excluded countries with unavailable or inconsistent data across periods from aggregate results. At the individual level, we drop respondents having a missing value in any indicator. The EU SILC data for the retained sample are then adjusted for missing observations using sampling weights. By reweighted the retained sample, we maintain the original population of each country.9 For countries with registry data the measure is constructed only from respondents with information in all indicators and using the specific sampling weight for this subgroup (PB060). The proportion of the sample with registry data is given on the far right column. 8 Based on Geodeme (2010) whose do file is available at and AnneCatherine Guio s comments. 9 Our bias analyses do not show significant differences between the remaining population and the individuals excluded due to nonresponse or missing variables in all countries. To test for bias when there are large (15% or more) drops in sample size we compare the uncensored headcount ratios for the retained and dropped groups of the population, and conduct a t test for differences between means following Alkire and Santos For more information on the use of register data in the the EUSILC survey, see Jäntti, Törmälehto and Marlier (2013). 11

14 Table 1: NonResponse and Missing Values across deprivation indicators reg. data AT 0.25% 0.13% 0.04% 0.13% 0.13% 0.07% 0.04% 0.00% BE 6.79% 10.27% 5.15% 2.07% 2.53% 2.60% 0.00% BG 17.81% 1.85% 1.49% 1.33% 1.16% 1.60% 0.00% CH 1.02% 0.74% 0.61% 0.73% 0.89% 0.00% CY 1.60% 1.53% 1.63% 1.30% 1.12% 1.14% 0.81% 0.00% CZ 10.00% 10.47% 14.05% 16.80% 20.93% 26.02% 24.55% 0.00% DE 4.11% 4.69% 8.16% 6.12% 7.04% 8.88% 8.42% 0.00% DK 5.21% 4.95% 4.62% 8.44% 4.22% 2.56% 2.28% 37.30% EE 0.35% 0.93% 11.56% 19.96% 21.59% 19.69% 21.00% 0.00% EL 2.89% 2.89% 2.77% 2.10% 2.15% 1.68% 1.65% 0.00% ES 6.23% 5.78% 5.96% 5.70% 5.63% 6.13% 5.16% 0.00% FI 15.17% 14.58% 3.54% 3.71% 4.81% 4.61% 7.36% 41.60% FR 1.38% 6.44% 1.59% 1.63% 1.60% 1.67% 1.83% 0.00% HR 44.13% 44.34% 0.00% HU 0.29% 0.49% 0.43% 3.70% 1.53% 0.42% 0.61% 0.00% IE 2.10% 1.87% 1.96% 3.15% 12.01% 3.80% 0.00% IS 1.36% 1.48% 1.07% 1.37% 1.56% 1.70% 1.54% 47.27% IT 0.35% 4.57% 4.83% 4.41% 3.97% 6.08% 5.27% 0.00% LT 2.18% 2.01% 19.58% 15.92% 16.10% 14.49% 18.31% 0.00% LU 2.02% 1.20% 0.87% 1.89% 2.29% 3.05% 1.87% 0.00% LV 0.29% 0.18% 0.25% 1.59% 1.32% 1.17% 1.22% 0.00% MT 2.29% 1.63% 1.74% 1.39% 1.01% 0.00% NL 2.53% 2.00% 2.11% 1.99% 1.82% 1.97% 1.81% 44.44% NO 2.77% 3.30% 4.06% 5.18% 3.53% 6.12% 2.77% 39.12% PL 0.24% 0.27% 0.24% 0.17% 0.19% 8.25% 7.36% 0.00% PT 13.22% 12.75% 12.31% 11.37% 11.06% 10.87% 12.16% 0.00% 12

15 RO 4.30% 2.21% 1.98% 1.75% 3.74% 3.59% 0.00% SE 10.45% 10.02% 3.06% 2.57% 3.04% 2.93% 4.97% 40.72% SI 0.39% 0.14% 0.70% 0.63% 0.57% 0.55% 0.42% 55.72% SK 1.26% 0.74% 5.84% 4.20% 3.27% 3.64% 2.35% 0.00% UK 11.19% 10.27% 15.95% 18.13% 18.20% 19.83% 12.21% 0.00% 3.4 Uncensored headcount ratios of deprivations in each indicator The deprivations in all indicators in the years 2006 and 2012 are reported in Table 4 below. The table includes all deprivations of all individuals for whom no data on any indicator is missing. There are several points to note. First, the AROP percentages roughly match those published in other sources (Nolan et al 2010). 10 Second, in the aggregate data, of the three indicators used in the EU2020 poverty measure, deprivations in income (15.3% in the selected EUSILC countries) are the highest although this varies by country. (Quasi) joblessness and severe material deprivation tend to be lower and are, on average, 10.1% and 7.7%, respectively. The indicators that tend to have the highest incidence overall are perceptual data of chronic health status, and the selfreported incidence of noise or housing problems. However incidence varies considerably across countries. The challenges inherent in interpreting the subjective indicator levels and trends are biases from personality and adaptive preferences or knowledge asymmetries that may evolve over time. The fact that these indicators carry a lighter weight may ease interpretation of the trends somewhat. In education we merely remind the reader that educational deprivations depend in part upon the definition of primary school, and the duration thereof varies across the included countries. Finally, across the selfreported environmental indicators we see less variation overall than in other indicators which raises questions about whether they reflect shifting aspirations. Several empirical techniques could be useful to understand the interrelationships between indicators. For these illustrative measures, rather than provide a full justification as in Guio et al or Whelan et al (to which we refer readers), we present correlations and redundancy measures across the binary deprivation indicators using the pooled data across all persons in all years. 10 We are grateful to Brian Nolan and Bernard Maitre for direction in constructing this variable. 13

16 In the case of binary deprivation indicators, correlations generate Cramer s V. There are high intracorrelations among the health indicators (which are addressed below) and otherwise relatively low correlations. Low associations are desirable when a measure seeks to bring into focus multiple aspects of poverty; however the robustness of such measures to changes in weights must be ascertained, and also correlation analysis may not be the most precise tool (Alkire et al. 2015). Cramer s V results are presented in Table 2. 14

17 Table 1: Uncensored Headcount ratios of all indicators, 2006 & 2012 AROP qjobless sev. mat dep Education noise pollution crime housing health chr. illness morbidity AT 12.1%13.7%* 7.6%7.2% 3.5%3.7% 1.1%0.9% 18.8%19.9% 7.6%11%** 12%11.8% 10.1%11% 7.9%9%* 21.9%33%** 9.4%9.5% CY 16.9%14.8%* 4.7%7.1%** 12.8%14.5%* 24.9%20.7%** 36.5%26.6%** 24.8%15.7%** 12.7%15.4%** 35.3%30.5%** 9.4%6.7%** 29%32.4%** 8.5%7.8% CZ 8.5%8.9% 8.5%6.3%** 9.1%6.4%** 0.1%0.1% 18.8%14.2%** 19.3%15.4%** 14%13.1% 20.5%10.2%** 13.4%12.8% 29.8%30% 6.8%6.2%* DE 12.3%15.3%** 10.7%7.7%** 4.6%4.6% 3.4%2.9%* 27.2%25.9%* 23.7%21.9%** 12.4%11.8% 14%12.7%* 9.3%8.6%* 38.2%36.9%* 8.2%10.9%** DK 12.2%13.8%* 8.4%9.8%* 2.9%2.7% 0.2%0.1% 18.8%17.8% 7.6%5.8%** 13.4%10.5%** 8.3%16.9%** 7.8%7.6% 29.6%29% 0%6.8%** EE 18%17.8% 6.9%9.2%** 7.1%9.4%** 5.4%3.1%** 22.8%12.9%** 21.5%11.8%** 20.2%15.6%** 23.3%19.3%** 15%16.3%* 38.5%43.6%** 9.5%9.8% EL 20.3%22.7%* 21.2%15.2%** 20.4%25.1%** 17.1%26.2%** 8.4%19.8%** 35.7%26.4%** 9.2%9.3% 20.2%23.7%* 6%10%** 11.9%19.6%** 23.1%31%** ES 19.2%21%* 7.5%14.7%** 3.2%5.5%** 31.4%24.8%** 27%14.9%** 17.1%7.9%** 19.9%10.3%** 17.1%12%** 12.2%8%** 23.8%26.1%** 8.6%5.1%** FR 12.6%12.9% 8.8%7.7%* 4.8%4.9% 23.9%17.2%** 19.6%16.8%** 15.8%11.1%** 16.1%14.6%* 11.6%12.2% 9.5%8.5%** 34.3%36.5%** 6.3%8.7%** HU 13.9%12.5% 11.9%10.6%* 20.1%24.3%** 8.1%3.5%** 17.2%10.1%** 13.1%11.8%* 9.9%10.1% 26.7%23.2%** 20.3%15.9%** 35.6%35.8% 13.4%7.8%** IS 9%7.1%** 3.2%5.8%** 1.9%2.2% 3.5%2.2%** 12.4%11.2% 8.2%8.4% 2.4%3.3%* 12.1%17.2%** 4.7%5% 24.4%28.8%** 5.7%10.1%** IT 18.7%18.2% 11.4%10.8% 6.2%14%** 26.7%20.5%** 25.1%17.7%** 21.6%17.1%** 14.9%14.8% 21.9%21.2% 10.5%12.4%** 21.5%24.5%** 7%9.4%** LT 19.8%18.4% 8.7%10.9%* 26.2%20.4%** 9.6%9.4% 19.9%13.2%** 14.1%14.7% 7.7%4.9%** 28.1%17.7%** 18.1%20.4%** 33.4%29.5%** 10.2%8.1%** LU 12.8%13.5% 5.7%6.5% 0.9%1.2% 29%24.7%** 23%16.5%** 17.8%13.3%** 11.5%14.1%* 14.2%16.2%* 7.3%7.3% 23.6%20.1%** 7%5.8%* LV 22.6%18.4%** 7%12%** 30.8%25.4%** 3.7%2.8%** 20.8%15.7%** 32.1%21.9%** 25.7%16.9%** 31.7%28%** 19.4%15.2%** 35.1%35.8% 10.1%7%** NL 9%9.6% 10.5%7.7%** 2%2% 10.8%8.3%** 31.4%24.2%** 14.2%13.8% 16.3%18.2%* 16%15.7% 5.2%5.7% 32%34.5%** 8.2%5.8%** NO 12.1%10.7%* 8.1%6.8%* 2.9%1.7%** 0.2%0.3% 12.5%11% 7.6%9.6%* 4%5.9%** 7.6%7.7% 9.4%6.3%** 33.7%30.8%* 8.7%4.3%** PL 17.8%16.3%* 14.3%11.8%** 27.8%13.5%** 20.4%16.4%** 19.8%14.4%** 12.9%11%** 8.9%6.4%** 41.2%10.4%** 17.3%14.6%** 32.4%34.4%** 6.2%7.4%** PT 18.1%17.2% 7.2%10.5%** 8.9%8.3% 54.5%47.2%** 25.5%23.8% 20.4%15.4%** 12%11% 19%22%* 20%18.2%* 30.8%37%** 11.5%21.8%** SE 11.9%14.3%** 6.2%9%** 1.9%1.3%* 9.6%8.3%** 13%13.1% 6.9%7.9%* 13.1%9.5%** 6.6%7.2% 5.8%4.2%** 35%33.8% 8.2%6.1%** SI 11.6%13.5%** 8%8.5% 5.3%6.8%** 21.4%3.1%** 18%14.1%** 20.6%16.2%** 9.5%8.2%** 21.8%31.6%** 15.7%12.4%** 36.4%35.3% 8.4%11.4%** 15

18 SK 10.7%11.9%* 7.1%7.2% 17.8%10.3%** 1.4%0.6%** 19.8%16%** 19.8%15.3%** 8.3%9.7%* 6.3%8.4%** 18%12.5%** 27.4%29.7%** 11.1%10%* UK 18%15.7%** 13.6%9.8%** 3.9%6.9%** 0%0% 22%17.8%** 13.4%8.2%** 27.6%19.5%** 13.2%16%** 6.5%8.2%** 38.1%32.8%** 8.5%10.6%** Aggregate 15.3%15.6% 10.5%9.7%** 7.5%7.8% 15.1%11.6%** 23.3%18.5%** 17.4%13.6%** 15.8%13.1%** 17.9%14.6%** 10.8%9.9%** 31.8%32.4%** 7.9%9%** Countries with missing years BE: %14.6% 14.1%20.5%** 22.5%19.1%** 15.7%16.3% 17.9%15.4%* 15%14.3%* 8.4%9.6% 24.7%26.1%* 7.8%8.4%* 5.7%5.1%* 31.8%31.5% FI: %13.7% 4.3%5.8%* 16.4%14.3%* 12.8%8.9%** 15%8.6%** 0%0% 9.9%6.8%* 43.1%46.7% 12.1%7.2%** 3.4%3% 26.9%29.1%* IE: %14.7% 14.6%10.9%** 14.3%9.2%** 8.5%4%** 16%10.4%** 21.9%17.6%** 3.2%2.9% 25.4%26.3% 6.1%4.9%* 4.1%7.2%* 20.9%29.1%** BG: %20% 13.7%13% 15.4%11.9%* 24.2%14.4%** 27.4%26.9% 10%6.2%** 16.7%11.8%** 28.7%18.6%** 2.4%3.9%* 57.9%43.7%* 28.2%23.8%** RO: %21.1%* 29.1%14.8%** 34.6%26.9%** 18.4%17.5% 14.7%13.6%* 10.8%8.4%* 10.2%9.5% 20.1%20% 7.4%8% 36.4%28.5%** 21.1%21.8% CH: %15.6% 7.5%11.9%* 18.4%19.2%* 13%10.3%* 12.6%16.9%** 18.7%16.3%* 3.2%3.1% 31.7%34%** 5.2%5.7% 1.9%0.8% 20.1%21.4%* MT: %13.4%** 6.9%11%** 25.1%30.6%** 36.6%40.5%** 9.9%12%** 22.1%21.5% 4.3%3.4% 24.5%28.9%* 2.6%2.8% 3.7%7.6%* 21.8%21.3% HR: %20.2%* 15.1%13.7%* 11.1%10.2% 7%7.1% 3.4%3.2% 9.2%7.2%* 26.9%25.7%* 36.9%29.3%** 7.8%5.2%* 14.9%15% 30.3%30.5% 16

19 qjobless Table 3: Correlations (Cramers V) across uncensored headcount ratios sev. mat chr. education noise pollution crime housing health dep illness morbidity u.m. needs AROP qjobless sev. mat dep education noise pollution crime housing health chr. illness morbidity u.m. needs 1.00 Table 3 presents a measure of Redundancy which is a more precise assessment for our purposes. It draws on the crosstabulation of the dichotomised deprivation status of persons in each indicator (Alkire et al 2015, 7.2). The redundancy value is the percentage of the population experiencing both deprivations, divided by the lower of the two marginal headcount ratios of deprivation. Each redundancy value shows the percentage of people who are deprived in both indicators as proportion of those deprived in the one with the lower headcount, hence it ranges from 0 to 100%. For example: in the case of (quasi) joblessness and atriskof[income]poverty, only 27% of the people who are quasijobless are also atriskofincomepoverty. The highest redundancy of 55% is between morbidity and health that is, 55% of those who are deprived in terms of morbidity have low selfreported health. Redundancy complements correlation analysis in important ways: despite the fact that morbidity and selfreported health have a high correlation of over 0.9, this more precise redundancy indicator shows that in 45% of cases, persons who report deprivations in morbidity do not experience low selfreported health, and for this reason, both variables are retained. 17

20 Table 4: Redundancy values across uncensored headcount ratios q jobless sev. mat dep education noise pollution crime housing health chr. illness morbidity u.m. needs AROP qjobless sev. mat dep education noise pollution crime housing health chr. illness morbidity u.m. needs Definition of Experimental Measures: Dimensions, Indicators, Cutoffs and Weights Having described the deprivations, we now set out the experimental measures that are implemented as described in Table 5. As mentioned above, three measures are constructed with varied weighting structures. The measures are computed and reported for all available time periods to analyse changes across time. All measures include the same 12 indicators. Three are indicators of the EU2020 multidimensional poverty index: income poverty (framed as being atriskofpoverty AROP); severe material deprivation; and (quasi) joblessness. Health has four indicators: selfreported health, the presence of a chronic illness, activity limitations due to poor health and unmet medical needs. Living Environment has four indicators: housing, pollution, crime and noise. Table 5 explains the indicators and cutoffs, and measurespecific weights. In Measure 1, the 12 indicators are organized into four dimensions. Each dimension is equally weighted, and each indicator within a dimension is equally weighted. Measure 2 replicates Measure 1 except that the indicator of material deprivation is consider a fifth dimension. Measure 3 organises the 18

21 12 indicators into 6 dimensions by breaking up (and effectively trebling the weight on) the EU2020 indicators, which are each considered separate dimensions. In Measure 3, the weight on AROP, (quasi) joblessness and severe material deprivation are 1/6 each, as are the weights on health, education and environment. Table 5: Dimensions, Indicators and Weights for Measures (M) 1, 2 and 3 Dimension Variable Respondent is not deprived if: M1 M2 M3 EU 2020 AROP The respondent s equivalized disposable income is above 60 per cent of the national median 1/12 1/10 1/6 QuasiJoblessness The respondent lives in household where the ratio of the total number of months that all household members aged 1659 have worked during the income reference year and the total number of months the same household members theoretically could have worked in the same period is higher than 0.2 1/12 1/10 1/6 Severe material deprivation The respondent has at least six of the following: the ability to make ends meet; to afford one week of holidays; a meal with meat, chicken, fish or vegi equivalent; to face unexpected expenses; and, to keep home adequately warm. Or the respondent has a car, a colour TV, a washing machine, and a telephone. 1/12 1/5 1/6 Education Education The respondent has completed primary education 1/4 1/5 1/6 Environment Noise The respondent lives in a household with low noise from neighbourhood or from the street 1/16 1/20 1/24 Pollution The respondent lives in a household with low pollution, grime or other environmental problems 1/16 1/20 1/24 Crime The respondent lives in a household with low crime, violence or vandalism in the area 1/16 1/20 1/24 Housing The respondent lives in a household with no leaking roof, damp walls, rot in window frames of floor 1/16 1/20 1/24 Health Health The respondent considers her own health as fair or above 1/16 1/20 1/24 Chronic Illness The respondent has no chronic illness or longterm condition 1/16 1/20 1/24 Morbidity The respondent has no limitations due to health problems 1/16 1/20 1/24 Unmet Med. Needs The respondent does not report unmet medical needs 1/16 1/20 1/24 19

22 4. Results This section presents the results for three measures across seven periods. 4.1 Identification of multidimensional poverty The first step is to identify who is poor. The AF dual cutoff methodology identifies a person as poor if the weighted sum of his or her deprivations is greater than or equal to the poverty cutoff. It censors the deprivations of the nonpoor, in order to focus attention strictly on the poor. Having identified the poor, the methodology then aggregates information regarding the poor into an overall poverty measure. We first calculate the poverty measure using ten poverty cutoffs at 10% increments, for all measures in all periods Using Measure 1 and data of three years: 2006, 2009 and 2012, Figure 1 compares the level of poverty of four geographic regions according to United Nations definitions11. Clearly, Northern and Western Europe are significantly the two least poor regions (respectively) regardless the year and cutoff. Southern Europe is the poorest region up to the 40% cut off. At 50% and more, differences between Eastern and Southern Europe are not significant United Nations classify Cyprus as Western Asia; however, we included it into Southern Europe as otherwise Cyprus would have been excluded. 20

23 Figure 1: Measure 1 Adjusted Headcount Ratio (M 0 ) by poverty cutoff !0.18!! M0#!0.18!! M0#!0.18!! M0#!0.16!! Eastern! Europe!!0.16!! Eastern! Europe!!0.16!! Eastern! Europe!!0.14!!!0.12!! Northern! Europe!!0.14!!!0.12!! Northern! Europe!!0.14!!!0.12!! Northern! Europe!!0.10!!!0.08!! Southern! Europe!!0.10!!!0.08!! Southern! Europe!!0.10!!!0.08!! Southern! Europe!!0.06!! Western! Europe!!0.06!! Western! Europe!!0.06!! Western! Europe!!0.04!!!0.04!!!0.04!!!0.02!!!0.02!!!0.02!!!"!! k#!"!! k#!"!! k# Figure 2 analyses the pooled information of countries over time including only countries with consistent and available information across years (as listed in the beginning of Section 3). Measures 1 and 2 show a significant reduction in multidimensional poverty in Europe between 2006 and 2012 (for cutoffs of 60% and higher, this difference is not significant); dominance is not clear between 2006 and For instance, in Measure 1 and 2, poverty reduction is only statistically significant with a cutoff of 10%. In Measure 3, there is not clear dominance for any poverty cutoffs above 40% in any pair of consecutive years, although for cutoffs 40% and below, poverty in 2012 is significantly lower than in other years. 21

24 Figure 2: Measure 1,2 & 3: Adjusted Headcount Ratio (M 0 ) by poverty cutoff Measure 1 Measure 2 Measure # M0# 0.16! M0# 0.16! M0# 2006! 2006! 0.14# 2009# 0.14! 2009! 0.14! 2009! 2012! 2012! 0.12# 0.12! 0.12! 0.10# 0.10! 0.10! 0.08# 0.08! 0.08! 0.06# 0.06! 0.06! 0.04# 0.04! 0.04! 0.02# 0.02! 0.02! k# k# k# 0.00# 10%# 20%# 30%# 40%# 50%# 60%# 70%# 80%# 90%#100%# 0.00! 10%!20%!30%!40%!50%!60%!70%!80%!90%!100%! 0.00! 10%!20%!30%!40%!50%!60%!70%!80%!90%!100%! In what follows we have selected poverty cutoffs which require a person to be poor in strictly greater than one dimension or the equivalent sum of weighted deprivations drawn from several dimensions. This definition assures that each person identified as poor is indeed deprived in two or more dimensions, which coheres with popularion understandings of multidimensional poverty. 12 Table 6 presents results for all Measures in 2006, 2009 and The poverty cutoffs of 26% and 21% used in Measures 1 and 2 identify a person as multidimensionally poor if they are deprived in strictly more than one dimension, or in some equivalent set of weighted indicators. Measure 3 s poverty cutoff of 34% identifies a person as multidimensionally poor if they are deprived in strictly more than two dimensions or the equivalent weighted indicators. 12 We are grateful to Tony Atkinson for suggesting that this conceptual issue needs to be addressed and, when the purpose of the measure permits, satisfied. 22

25 Table 6: Aggregate Results and consistent partial indices Measure 1 k=26% Measure 2 k=21% Measure 3 k=34% Adjusted Headcount Ratio (M 0 ) Headcount Ratio (H) 19.6% 17.6% 16.1% 22.3% 19.9% 19.0% 9.8% 8.4% 8.4% Intensity (A) 40.1% 39.6% 39.5% 36.5% 35.7% 36.4% 48.2% 47.8% 48.5% Censored Headcount ratios: AROP 7.4% 6.8% 6.5% 8.7% 8.1% 8.1% 6.9% 6.2% 6.2% (Quasi)Joblessness 5.5% 4.7% 4.7% 6.5% 5.6% 5.8% 5.3% 4.6% 4.8% Severe material deprivation 4.8% 3.6% 4.4% 6.9% 5.4% 6.7% 4.5% 3.6% 4.3% Education 12.3% 11.0% 9.4% 12.3% 11.0% 9.4% 5.0% 4.1% 3.7% Noise 7.7% 6.7% 5.6% 8.1% 7.1% 6.0% 3.7% 3.2% 2.8% Pollution 6.1% 5.3% 4.3% 6.2% 5.5% 4.5% 2.7% 2.4% 2.1% Crime 5.4% 4.9% 4.2% 5.6% 5.2% 4.5% 2.7% 2.4% 2.1% Housing 7.3% 5.5% 5.1% 8.2% 6.1% 5.8% 4.4% 3.1% 3.1% Health 6.8% 5.9% 5.7% 6.9% 6.0% 5.8% 3.4% 2.8% 2.7% Chronic Illness 11.8% 11.3% 10.7% 12.4% 11.9% 11.3% 5.5% 4.9% 4.9% Morbidity 4.7% 4.7% 4.9% 4.7% 4.7% 4.9% 2.2% 2.1% 2.2% Unmet Med. Needs 3.2% 2.7% 2.6% 3.5% 2.9% 2.8% 1.9% 1.5% 1.6% Percentage Contributions (weighted) AROP 7.9% 8.2% 8.6% 10.7% 11.4% 11.8% 24.2% 25.7% 25.4% (Quasi)Joblessness 5.9% 5.6% 6.2% 7.9% 7.9% 8.3% 18.7% 19.0% 19.6% Severe material deprivations 5.0% 4.4% 5.8% 16.9% 15.2% 19.5% 16.0% 14.8% 17.7% Education 39.1% 39.4% 37.1% 30.2% 30.8% 27.3% 17.5% 17.0% 15.3% Noise 6.1% 6.0% 5.5% 5.0% 5.0% 4.3% 3.3% 3.4% 2.9% Pollution 4.8% 4.7% 4.3% 3.8% 3.9% 3.3% 2.4% 2.5% 2.1% Crime 4.3% 4.4% 4.1% 3.5% 3.7% 3.2% 2.4% 2.5% 2.2% 23

26 Housing 5.8% 5.0% 5.0% 5.1% 4.3% 4.2% 3.9% 3.3% 3.2% Health 5.4% 5.3% 5.6% 4.2% 4.2% 4.2% 3.0% 2.9% 2.7% Chronic Illness 9.4% 10.2% 10.5% 7.6% 8.4% 8.2% 4.9% 5.1% 5.0% Morbidity 3.7% 4.2% 4.8% 2.9% 3.3% 3.5% 2.0% 2.2% 2.2% Unmet Med. Needs 2.5% 2.5% 2.6% 2.1% 2.1% 2.1% 1.7% 1.6% 1.6% We see that the level of poverty is lowest in Measure 3, followed by Measure 1 and then Measure 2. This is likely to reflect the poverty cutoffs, which are highest in Measure 3 and lowest in Measure 2. The identification of who is poor in each measure is different, and reflects both the poverty cutoffs and the weights, and we see a markedly lower level of deprivation in Education, Environment and Health indicators for Measure 3, where these dimensions have a lower weight of 1/6. Figure 3: Headcount ratio and intensity SILC selected countries Measure 1 k=26% Measure 2 k=21% Measure 3 k=34% 41%! A# 38%! A# 50%! A# 40%! 39%! 2012! 2009! 2006! 37%! 36%! 2012! 2009! 2006! 49%! 48%! 2012! 2009! 2006! 38%! H# 15%! 16%! 17%! 18%! 19%! 20%! 35%! H# 18%! 19%! 20%! 21%! 22%! 23%! 47%! H# 6%! 7%! 8%! 9%! 10%! 11%! Across years and measures, the reduction in the level of multidimensional poverty clearly reflects the reduction in the percentage of poor individuals as seen in figure 3. Each point represents a multidimensional poverty level base on the headcount ratio (horizontal axis) and the intensity (vertical axis). Each additional point shows confidence intervals for the headcount ratio and intensity. In Measure 1, the headcount ratio was 19.4% in 2006 and fell significantly to 16.1% in The intensity of poverty in the pooled sample dropped marginally but significantly by 0.05%, but with large differences across countries. Measure 2 and 3 show significant reductions in the headcount ratio and intensity between 2006 and Between 2009 and 2012, Measure 2 displays a small but significant reduction in the headcount (from 19.9% to 19.0%) and an increase in the intensity reaching a level similar to During the same period, Measure 3 presents a marginal rise in the headcount ratio and an insignificant increase in the intensity from 47.8% to 48.5%. 24

27 Figure 4: Dimensional Breakdown SILC selected countries Model$3$k=34%$ Model$2$k=21%$ Model$1$k=26%$ 2012$ 2009$ 2006$ 2012$ 2009$ 2006$ 2012$ 2009$ 2006$ 9%$ 8%$ 8%$ 12%$ 11%$ 11%$ 6%$ 6%$ 6%$ 4%$ 6%$ 5%$ 8%$ 8%$ 8%$ 25%$ 26%$ 24%$ 37%$ 39%$ 39%$ 20%$ 15%$ 17%$ 20%$ 19%$ 19%$ 27%$ 31%$ 30%$ 18%$ 15%$ 16%$ 5%$ 6%$ 6%$ 4%$ 4%$ 5%$ 6%$ 11%$ 5%$ 3%$ 5%$ 4%$ 5%$ 4%$ 5%$ 6%$ 5%$ 5%$ 4%$ 3%$ 3%$ 4%$ 4%$ 5%$ 4%$ 4%$ 4%$ 4%$ 5%$ 4%$ 3%$ 5%$ 4%$ 10%$ 9%$ 8%$ 8%$ 8%$ 4%$ 2%$ 4%$ 3%$ 4%$ 2%$ 3%$ 2%$ 3%$ 2%$ 15%$ 3%$ 2%$ 2%$ 3%$ 3%$ 5%$ 2%$ 2%$ 17%$ 3%$ 3%$ 3%$ 3%$ 3%$ 5%$ 2%$ 2%$ 18%$ 3%$ 2%$ 2%$ 4%$ 3%$ 5%$ 2%$ 2%$ AROP$ q9jobless$ sev.$mat$dep$$ educaeon$ noise$ pollueon$ crime$ housing$ roof$ health$ chr.$illness$ morbidity$ u.m.$needs$ 0%$ 10%$ 20%$ 30%$ 40%$ 50%$ 60%$ 70%$ 80%$ 90%$ 100%$ The percentage contributions of each indicator to overall poverty reflect the weights and the censored headcounts. We see a tremendous difference here: in Measure 1, the three EU2020 indicators together contribute around 1821% to poverty (less than their joint weight of 25%) whereas in Measure 3, they contribute 5963% to poverty (greater than their joint weight of 50%). Education contributes 39% in Measure 1, but only 1517% in Measure 3, and health and environment both contribute 1924% in Measure 1, and 1012% in Measure 3. Thus the measures, as designed, do indeed illuminate different deprivation profiles according to their composition. 4.2 Poverty across countries This section presents and discusses the three measures results in the year To make comparisons we use the previously mentioned poverty cutoff for each measure. For each measure, we present the M 0 value as well as its associated partial indices (H) and (A):. From Table 7, we see first of all that each of the three measures in 2009, which differ in weights and indicators, generate relatively similar country rankings. Kendal tau b rank correlations across the countries range from 0.69 (between Measures 1 and 3) to 0.85 (between Measures 2 and 3), which 25

28 suggest that more detailed robustness tests may find the measures to be relatively robust in indicators and weights. 13 Table 7: Aggregated Results by Measure and country in 2009 NO IS DK FI AT CZ UK SE NL SK DE SI CH EE BE LU IE FR HU MT LT ES Measure 1 k=26% Measure 2 k=21% Measure 3 k=34% M0 H A M0 H A M0 H A % 34.2% % 33.9% % 46.1% % 34.7% % 30.9% % 45.0% % 34.6% % 33.0% % 45.9% % 33.9% % 33.6% % 46.4% % 36.2% % 35.0% % 48.3% % 34.7% % 34.4% % 47.6% % 34.3% % 32.9% % 45.5% % 37.0% % 32.6% % 44.9% % 37.2% % 32.7% % 46.6% % 35.4% % 35.2% % 47.9% % 37.8% % 35.8% % 48.9% % 37.8% % 35.3% % 47.9% % 37.8% % 33.0% % 44.9% % 36.9% % 35.3% % 47.9% % 40.5% % 36.9% % 49.2% % 39.1% % 33.3% % 45.3% % 39.7% % 35.8% % 46.7% % 39.8% % 34.8% % 47.8% % 38.7% % 36.9% % 48.8% % 40.6% % 35.4% % 47.5% % 39.9% % 37.6% % 47.7% % 39.6% % 34.4% % 46.0% 13 The Kendall Taub between Measures 1 and 2 is For assessments of robustness to weights and cutoffs see Alkire and Santos 2014, Alkire et al. 2015; Ura et al

29 IT PL CY LV EL RO BG PT % 41.3% % 36.6% % 47.8% % 42.4% % 39.3% % 49.7% % 42.6% % 37.5% % 47.1% % 38.2% % 38.4% % 49.1% % 41.9% % 37.7% % 48.5% % 40.2% % 38.2% % 48.1% % 40.3% % 39.5% % 50.4% % 41.9% % 36.7% % 48.1% As before, the levels of poverty provided by Measure 1 tend to be the highest, followed by Measure 2 and 3. In Measures 1 and 2, Portugal has the highest poverty rates; and Norway followed by Iceland and Denmark has the lowest poverty rates. In Measure 3, Bulgaria is the poorest and Iceland the least poor. Intensity varies considerably from 34% to 43% in Measure 1, and 31% to 40% in Measure 2 whereas in Measure 3 the range is only 45% to 50%. Intensity is not necessarily highest in the countries with highest poverty, a finding that contrasts with other studies. For example in Measure 2, Belgium with 20% of person s being poor, has a marginally higher point estimate of intensity than Portugal, where 41% of people are poor. Similarly in Measure 3, Austria in which 4.2% of people are poor has a higher point estimate of intensity than Portugal or Romania. After censoring the deprivations of nonpoor people, the Adjusted Headcount Ratio can be broken down by indicator. Figures 5, 6 and 7 provide the percentage contribution of each indicator of poverty Measure 1, 2 and 3 in the year 2009, respectively. The countries are ranked from those having highest rates of poverty to those with lowest rates. Measure 1 has four equally weighted dimensions. The percentage contribution of education varies greatly across countries and increases strikingly in the poorer countries. This reflects differences in achievements, but also in definitions of primary school, so unfortunately is not strictly comparable. The relative contribution of (quasi) joblessness declines as overall poverty in a country increases, as do the relative contributions of the health variables. In general, in the least poor countries the relative contribution of educational deprivations is lower and of EU2020 indicators (with some exceptions) is higher. This interesting finding draws attention to the need to consider noneu2020 indicators. The environmental indicators in the pink hues show relatively less variation across countries. 27

30 Figure 5: Dimensional Decomposition Measure 1 k=26% by country (2009) 0%! 10%! 20%! 30%! 40%! 50%! 60%! 70%! 80%! 90%! 100%! PT! EL! CY! ES! IT! PL! MT! BG! LU! LV! FR! RO! IE! BE! LT! CH! HU! EE! NL! DE! SI! SE! SK! UK! AT! CZ! FI! DK! IS! NO! AROP! q"jobless! sev.!mat!dep!! education! noise! pollution! crime! housing! health! chr.!illness! morbidity! u.m.!needs! 28

31 Figure 6: Dimensional Decomposition Measure 2 k=21% by country (2009) 0%! 10%! 20%! 30%! 40%! 50%! 60%! 70%! 80%! 90%! 100%! PT! BG! RO! EL! LV! PL! CY! IT! ES! LT! MT! HU! FR! IE! LU! BE! EE! CH! SI! DE! SK! NL! SE! UK! CZ! AT! FI! DK! IS! NO! AROP! q"jobless! sev.!mat!dep!! education! noise! pollution! crime! housing! health! chr.!illness! morbidity! u.m.!needs! 29

32 Figure 7: Dimensional Decomposition Measure 3 k=34% by country (2009) 0%! 10%! 20%! 30%! 40%! 50%! 60%! 70%! 80%! 90%! 100%! BG! LV! RO! PT! PL! EL! CY! LT! IE! IT! BE! ES! HU! MT! FR! EE! DE! LU! UK! SI! CH! SK! SE! NL! AT! FI! CZ! DK! NO! IS! AROP! q"jobless! sev.!mat!dep!! education! noise! pollution! crime! housing! health! chr.!illness! morbidity! u.m.!needs! In Measure 2, severe material deprivation and education lead the dimensional contribution to multidimensional poverty. In Measure 3, the EU2020 indicators AROP, severe material deprivation and (quasi) joblessness contribute more than 50% to poverty in all but three countries. Naturally the composition of poverty is affected both by the censored headcount ratios of each indicator and also by its weights. It can also therefore be useful to view the levels of deprivation in each indicator individually, separately from the weights. To do this we construct censored headcount ratios, which as mentioned previously show the percentage of people who are identified as poor and are deprived in each particular indicator. Note that the poverty measure M 0 is merely the weighted average of the censored headcount ratios that is, the sum of the censored headcount ratios of each indicator, where censored headcount is multiplied by its respective weight. Figure 8 below provides the uncensored and censored headcount ratios of indicators in three countries: Norway, Hungary and Portugal, using Measure 3 (k=34%) in The total height in light grey show the uncensored ( raw ) headcount ratios whereas the dark portion depicts the censored headcount ratios. Necessarily, the censored headcount ratios are equal to or lower than the 30

33 uncensored headcount ratios. The difference between these shows whether some persons who are deprived in that indicator are not simultaneously deprived in enough other indicators to be identified as multidimensionally poor The difference between uncensored and censored headcount ratios is particularly noticeable in chronic illness, health and housing as well as noise, crime and pollution. In this way the poverty cutoff may be used to clean the observations of deprivations that do not signify poverty in some cases because they may reflect varying frames of reference (noise), or standards (housing). Note also that the deprivations with the highest weight (AROP, (quasi) joblessness, education) have relatively less differences between uncensored and censored headcount ratios than the others because one requires fewer additional indicators to be identified as poor. Of these three, the differences between uncensored and censored headcount ratios in AROP tend to be larger, but this is not a fixed rule. Figure 8: Raw and Censored Headcount Ratios Measure 3 k=34% for Norway, Hungary and Portugal (2009) 50%! 45%! 40%! 35%! 30%! 25%! 20%! 15%! 11%!11%! 10%! 5%! 0%! Censored!Headcount! Uncensored!Headcount! 15%! 10%! 6%!7%! 19%! 2%! 7%! 5%! 49%! In Norway, 35% of the population is deprived in chronic illness. However, the dual cutoff approach shows that in many cases selfreported chronic illness is not associated with sufficient additional deprivations to identify a deprived person as poor. The percentage of the population who are poor in 34% or more of the weighted indicators, and who are deprived in chronic illness, is only 2.1%. A similar gap between uncensored and censored headcount ratio can be perceived in indicators like AROP and noise. Hungary and Portugal find a similar pattern for chronic illness as Norway. Over 30% of people are deprived in chronic illness, nevertheless only 6% and 9% of the population is poor and deprived in Hungary and Portugal, respectively. 0%! 13%!12%! 25%! 11%! 8%! 19%! 11%! 6%! 14%!14%! 8%! 19%!18%! 7%! 15%! 36%!35%! 30%! 8%! 6%! 8%!10%! 2%! 4%! HU! NO! PT! HU! NO! PT! HU! NO! PT! HU! NO! PT! HU! NO! PT! HU! NO! PT! HU! NO! PT! HU! NO! PT! HU! NO! PT! HU! NO! PT! HU! NO! PT! HU! NO! PT! AROP! q"jobless! sev.!mat! dep!! education! noise! pollution! crime! Housing! health! chr.!illness! morbidity! u.m.!needs! Hungary has a higher match between deprivations and poverty in AROP, (quasi) joblessness and education. The raw headcount of AROP, (quasi) joblessness and education deprivation is 11%, 10% 31

34 and 5%, respectively; and the censored headcount shows that 6% of the population is AROP and poor; meanwhile 5% is deprived in (quasi) joblessness and poor; and, 3% of the population is education and poor. Still, in each of these at least 40% of those who are deprived are not identified as poor. Finally, in Portugal, the gap between raw and censored headcount ratios is proportionally smallest for AROP, (quasi) joblessness and severe material deprivation, and largest for chronic illness, crime, noise and education. This section has illustrated the basic analyses of multidimensional poverty measures and their partial indices; the appended tables provide comprehensive results for all measures across all years, with varying poverty cutoffs. The next section analyses changes in multidimensional poverty across time. 4.3 Poverty across time: According to all measures, on average across all countries, multidimensional poverty measured by M 0 shows a significant decrease between 2006 and 2012 in absolute terms. Table 8 presents the absolute changes between 2006 and 2012 for each country and overall. There are significant absolute reductions in the headcount ratio in all measures between 2006 and However, intensity shows a significant change only in Measure 1. If we consider reductions in relative terms, Measure 1 shows the highest reduction (poverty reduced by 19% from its 2006 level, or by in absolute terms), followed by Measure 2 (15%0.012) and Measure 3 (13%0.006). In Measure 1, the poverty reduction occurs across both periods: and In Measure 2, the first triennium explains more than the 80% of the total reduction of poverty. Finally, in Measure 3, all poverty reduction is explained by changes in the period

35 Table 8: Aggregate Results by country 2006 and 2012, Measures 1, 2 and 3 Measure 1 k=26% Measure 2 k=21% Measure 3 k=34% M0 H A M0 H A M0 H A M0 H A M0 H A M0 H A AT % 35.1% % 35.2% % 33.3% % 33.8% % 47.4% % 46.6% CY % 44.3% % 41.7% % 39.2% % 37.1% % 48.7% % 46.9% CZ % 35.9% % 34.5% % 36.1% % 34.9% % 50.4% % 46.8% DE % 37.1% % 37.3% % 35.1% % 35.7% % 47.8% % 49.2% DK % 33.7% % 33.8% % 34.0% % 32.7% % 45.4% % 45.3% EE % 39.7% % 37.3% % 37.3% % 36.2% % 49.4% % 48.6% EL % 41.6% % 42.7% % 37.8% % 39.8% % 48.5% % 50.9% ES % 41.0% % 39.5% % 35.4% % 35.1% % 46.5% % 47.9% FR % 40.1% % 39.6% % 35.0% % 35.1% % 47.4% % 47.7% HU % 40.7% % 38.7% % 38.8% % 38.2% % 50.8% % 50.2% IS % 34.8% % 36.0% % 31.4% % 32.9% % 43.1% % 45.7% IT % 41.7% % 41.7% % 37.1% % 38.1% % 48.4% % 48.9% LT % 40.1% % 39.1% % 38.3% % 38.1% % 48.8% % 48.5% LU % 40.1% % 39.5% % 33.8% % 33.8% % 45.7% % 46.4% LV % 38.8% % 37.7% % 39.5% % 38.0% % 49.3% % 49.1% NL % 37.8% % 37.2% % 33.3% % 32.6% % 46.4% % 45.2% 33

36 NO % 34.0% % 33.9% % 32.9% % 33.2% % 45.8% % 44.9% PL % 42.6% % 41.6% % 40.6% % 38.4% % 50.1% % 50.1% PT % 41.6% % 42.8% % 36.4% % 37.5% % 48.6% % 48.4% SE % 36.4% % 36.3% % 31.9% % 32.3% % 45.1% % 43.9% SI % 40.8% % 37.6% % 36.2% % 35.6% % 48.4% % 48.2% SK % 35.2% % 36.0% % 34.9% % 36.1% % 47.3% % 50.0% UK % 34.6% % 34.5% % 33.6% % 35.2% % 46.4% % 47.8% Aggregate % 40.1% % 39.5% % 36.5% % 36.4% % 48.2% % 48.5% BE % 40.0% % 36.5% % 48.9% BG % 40.2% % 38.6% % 51.1% CH % 38.1% % 32.7% % 44.4% FI % 33.4% % 33.6% % 33.2% % 32.8% % 46.3% % 45.0% HR % 40.3% % 39.0% % 49.9% IE % 40.4% % 36.1% % 49.2% MT % 40.1% % 35.3% % 47.3% RO % 40.8% % 37.9% % 47.7% 34

37 Figure 9: Changes in the adjusted headcount ratio M 0 by region over time Measure 1 k=26% Measure 2 k=21% Measure 3 k=34% 0.35# M0# 0.35# M0# 0.35# M0# 0.30# 0.30# 0.30# 0.25# 0.25# 0.25# 0.20# 0.20# 0.20# 0.15# 0.15# 0.15# 0.10# 0.10# 0.10# 0.05# 0.05# 0.05# 0.00# 2007# 2008# 2009# 2010# 2011# 0.00# 2007# 2008# 2009# 2010# 2011# 0.00# 2007# 2008# 2009# 2010# 2011# Eastern#Europe# Northern#Europe# Southern#Europe# Western#Europe# Eastern#Europe# Northern#Europe# Southern#Europe# Western#Europe# Eastern#Europe# Northern#Europe# Southern#Europe# Western#Europe# Figure 9 shows the evolution of the adjusted headcount ratio M 0 between 2006 and 2012 for the European subregions. All measures show a reduction in the poverty level of Eastern Europe; however, this reduction is faster and significant during the first years. Southern Europe shows a parsimonious reduction till 2010, at which time it had the highest poverty in all measures. In Measure 1, changes in Southern Europe are not significant between consecutive years; however, they become significant when longer periods are analysed. In Measures 2 and 3, there is an apparent increase in poverty in Southern Europe from 2010 onwards, but these changes are not significant. Western Europe significantly reduces poverty during the first year in Measures 1 and 2 then small increments and finally a decline from In Measure 3, Western Europe does not show any significant change in any period but Finally, Northern Europe presents slight ups and downs during the first three years in all measures. Later, multidimensional poverty in the area only shows insignificant changes. Table 8 shows the evolution of all countries in all measures between 2006 and We see that three countries Poland, Latvia and Slovenia had the largest absolute reduction in poverty (M 0 ) according 35

38 to all three measures. France and Spain did the next best in terms of poverty reduction by Measure 1; in Measure 2 it was France and Lithuania, and in Measure 3 it was Cyprus and the Czech Republic. Table 9, 10 and 11 present absolute changes for all countries, years and models. Changes significant at 10% are marked with * in each cell and at 1% with **. Figure 10 below shows the evolution of M 0 across time for each measure across all countries. The empty gaps for some countries and years are due to the lack of comparable data. According to Measure 1, 19 countries which are 83% of the countries with comparable data (all except Austria, Denmark, Greece and Iceland, that is) experienced poverty reduction. Highest poverty reductions were seen in Poland and Slovakia from to and to 0.040, respectively. Measures 2 and 3 show a similar pattern. More than 72% of the countries experienced reductions in their poverty levels between 2006 and 2012 (77% or 17 countries and 73% or 16 countries in Measures 2 and 3, respectively). Unfortunately according to both measures, Austria, Denmark, Greece, Iceland and Italy have higher poverty levels in 2012 than in In Measure 1, the United Kingdom; and, in Measure 3, Portugal and Sweden also increased their poverty levels. In both measures the reduction is led by Poland. Most countries show low or no decrease from 2009 to In Measures 1 and 2, there is a relatively stronger decrease in poverty from Some of this apparent decrease may be due to drops in the (relative) AROP poverty rates due to the financial crisis, illustrating the need for care in interpreting mixed relative and absolute indicators. Patterns vary considerably by country. 36

39 Figure 10: Adjusted Headcount Ratio for all Measures by country ( ) Measure 1 k=26% Measure 2 k=21% Measure 3 k=34% 0.20# 0.18# 0.16# 0.14# PT# EL# 0.20# 0.18# 0.16# 0.14# BG# PT# EL# 0.20# 0.18# 0.16# 0.14# 0.12# 0.10# CY# 0.12# 0.10# 0.12# 0.10# BG# EL# 0.08# 0.08# 0.08# 0.06# 0.06# 0.06# 0.04# 0.04# 0.04# 0.02# 0.02# 0.02# 0.00# 2007#2008#2009#2010#2011# 0.00# 2007#2008#2009#2010#2011# 0.00# 2007#2008#2009#2010#2011# AT# BE# BG# CH# CY# CZ# DE# DK# EE# EL# ES# FI# FR# HR# HU# IE# IS# IT# LT# LU# LV# MT# NL# NO# PL# PT# RO# SE# SI# SK# UK# Portugal clearly has the highest level of poverty across time in the first two measures except in Measure 2 between 2006 and 2008 where Bulgaria was the poorest. In Portugal, poverty increased in 2006, then it decreased until 2010 not significantly between 2008 and 2010 to increase again in the last period. In Measure 1, Greece is consistently the second poorest country. Norway and Iceland were the least poor countries in Measures 2 and 3; however, changes were mainly insignificant (except for the last period in the case of Norway and in the case of Iceland). Across countries, Poland is the only country that consistently decreases poverty in all Measures and periods; however, this change is not significant in in Measures 1 and 2 and and in Measure 3. Bulgaria with Poland and Slovenia presented sharp poverty reductions from 2007 (data are not available for 2006) except in 2011 (Measures 1 and 2). Germany and UK, on the other hand, remain stable without significant changes in any period except Germany in Measure 2. In Measure 2, Spain displays a constant reduction in poverty during the first periods. In 2008, poverty increases marginally to decrease in the next period. None of these changes is significant. From

40 poverty falls but the decrease is significant only between 2010 and In Measure 2 and 3, changes in 2008 and 2012 are positive, but insignificant. Italy presented sustained poverty reduction till 2010 when the level sharply increases. Between, 2011 and 2012, the situation was ambiguous. Measures 1 suggests a reduction in poverty while Measures 2 and 3, on the contrary, report a new increment. France shows a single trend. Positive and negatives changes are interspersed in all periods and year before Austria, Belgium and Denmark seem to show the highest increase in poverty. Normally the poverty analyses are undertaken at the country level to facilitate national policy design. However it can be quite interesting from a humancentric perspective to look across countries, and see where the people who are identified as poor by each measure live, and what proportion of poverty each country contributes to the whole. Figure 11, below, provides this information. Among the 22 countries used in this analysis, we have aggregated their M 0 measures using annual population figures for each of the years 2006 to The height of the stripe associated with each country depicts that countries relative contribution to the overall M 0 of the 22 countries together. The graphic also depicts what was already seen earlier, namely the sharp drop between and the relative stability of poverty Due to their size, Italy, France, Spain, Poland and Germany dominate poverty trends in Europe. Italy reduces its relative contribution during the whole period but France s and Spain s contribution consistently falls only from Poland is the only country that reduces its relative contribution in all periods. Such depictures are useful complements to detailed national analyses. Furthermore, with changes in population share it is possible to decompose changes in multidimensional poverty that might arise from demographic shifts across countries. 38

41 Figure 11: Poverty contributions by country, populationweighted Measure # 0.08# 0.07# 0.06# 0.05# 0.04# 0.03# 0.02# 0.01# 0# SILC# SILC# SILC# SILC# SILC# EL# SILC# SILC# UK# EL# PT# EL# UK# EL# EL# UK# UK# PT# UK# EL# EL# DE# PT# PT# UK# PT# UK# DE# PT# DE# DE# PT# DE# PL# DE# DE# PL# PL# PL# PL# PL# PL# ES# ES# ES# ES# ES# ES# ES# FR# FR# FR# FR# FR# FR# FR# IT# IT# IT# IT# IT# IT# IT# 2007# 2008# 2009# 2010# 2011# CZ# LU# NO# EE# CY# DK# SI# SK# LV# AT# LT# SE# HU# NL# EL# UK# PT# DE# PL# ES# FR# IT# IS# SILC# The value of including the intensity in the poverty measure is evident in Figure 12 below. The bubble graphic plots the headcount and intensity of every country. The different periods are shown in contrasting colours. The size of the bubble corresponds to the population size of the country. We see, first of all, that across all countries and all periods, the intensity of poverty is highest in the countries which simultaneously have high headcount ratios of poverty located in the upper right hand corner. However we also see that at the same headcount ratio the intensities vary. We also see that in some countries the reduction of poverty does occur by reducing intensity strongly (e.g. Spain ). A measure focused solely on the reduction of the prevalence of poverty would overlook these important changes. Further, as was mentioned above, such a measure could not be broken down by indicator into consistent subindices (Alkire Foster and Santos 2011). 39

42 Figure 12: Bubble graph of changes Measure 1 by H and A %# 44%# Intensity# Poland# 2009# 42%# Italy# Portugal# 40%# France# 38%# Spain# 36%# Germany# 34%# UK# Headcount#Ratio# 32%# 0%# 5%# 10%# 15%# 20%# 25%# 30%# 35%# 40%# 45%# Poverty in Portugal on the right hand side increased between 2006 and Between 2006 and 2009, there was a reduction in the percentage on poor people and a slightly increase in their intensity. From 2009 to 2012, the change is mostly explained by the higher intensity. Poland reduced poverty mainly by reducing the headcount ratio and marginally by the intensity reduction. Italy decreased poverty by reducing the headcount ratio between 2006 and Between 2009 and 2012, the headcount ratio decreases only marginally and the intensity almost returned to 2006 levels. On the other hand, Spain shows two completely different patterns across years. In the first triennium poverty reductions were based mainly on intensity; and, from 2009 to 2012, these changes depended on reduction in the percentage of poor people. France displays a constant reduction in their poverty levels by reducing the headcount ratio, although the intensity remained nearly constant between periods. During the first triennium, Germany increased poverty because of the rise in intensity. The second triennium is characterized by the reduction of intensity and partially by the decline in the percentage of poor individuals. Finally, the United Kingdom decreases poverty mainly based on the headcount 40